GLM-Image Generator

Generate images from text prompts using the GLM-Image API.

Attribution: Based on glm-image by ViffyGwaanl (MIT License).

Setup

This skill supports two providers. You only need one.

Option A — GLM (BigModel / Zhipu AI)

Requires GLM_API_KEY from https://open.bigmodel.cn → Console → API Keys

export GLM_API_KEY=your-key
# or add to ~/.openclaw/config.json: { "api_key": "your-key" }
# or add GLM_API_KEY=your-key to .env

Option B — OpenRouter

Requires OPENROUTER_API_KEY from https://openrouter.ai → Keys

export OPENROUTER_API_KEY=your-key
# or add to ~/.openclaw/config.json: { "openrouter_api_key": "your-key" }
# or add OPENROUTER_API_KEY=your-key to .env

Default OpenRouter model: google/gemini-3-pro-image-preview. Other options: openai/gpt-5-image-mini, openai/gpt-5-image, google/gemini-2.5-flash-image-preview. See full list at https://openrouter.ai/collections/image-models

Auto-detection: if both keys are present, GLM is used. Override with --provider openrouter.

Usage

When a user requests image generation:

Step 0 — Verify at least one API key is configured

Run:

python3 -c "
import os, json, pathlib
glm = bool(os.environ.get('GLM_API_KEY'))
orouter = bool(os.environ.get('OPENROUTER_API_KEY'))
if not glm and not orouter:
    for p in ['~/.openclaw/config.json', '~/.claude/config.json']:
        try:
            d = json.loads(pathlib.Path(p).expanduser().read_text())
            if d.get('api_key'): glm = True
            if d.get('openrouter_api_key'): orouter = True
        except: pass
keys = []
if glm: keys.append('GLM_API_KEY')
if orouter: keys.append('OPENROUTER_API_KEY')
print('FOUND: ' + ', '.join(keys) if keys else 'KEY_MISSING')
"

If output is KEY_MISSING, tell the user:

"No API key is configured. This skill supports two providers — you only need one:

Option A — GLM (BigModel): Get key at https://open.bigmodel.cn → Console → API Keys, then:

export GLM_API_KEY=your-key

Option B — OpenRouter: Get key at https://openrouter.ai → Keys, then:

export OPENROUTER_API_KEY=your-key

For either option you can also add the key to ~/.openclaw/config.json:

{ "api_key": "glm-key" }
{ "openrouter_api_key": "openrouter-key" }
```"

Do not proceed until the user confirms a key is set.

Step 1 — Ask for language (MANDATORY, no exceptions)

Before running anything, ask:

"What language is your prompt in? Please choose: zh (Chinese), en (English), ja (Japanese), ko (Korean), fr (French), de (German), es (Spanish)."

Do NOT infer language from the user's message language or any other signal. Do NOT default to any language. Do NOT proceed until the user explicitly states a language code.

Step 2 — Run the generation script

python3 scripts/generate.py "<prompt>" --language <code>

--language is required. The script will error if omitted.

Other defaults:

• Size: 1088x1920 (portrait HD)
• Output: output/ folder
• No watermark

Step 3 — Return the result

Display the markdown image link and local file path.

Generate Image

python3 scripts/generate.py "<prompt>" --language <zh|en|ja|ko|fr|de|es>

Provider is auto-detected from available keys. Override explicitly:

# Force OpenRouter with a specific model
python3 scripts/generate.py "<prompt>" --language en --provider openrouter --model google/gemini-2.5-flash-image-preview

# Force GLM
python3 scripts/generate.py "<prompt>" --language zh --provider glm

Options

• --language: (Required) Prompt language. Must be explicitly provided by the user. Supported: zh (Chinese), en (English), ja (Japanese), ko (Korean), fr (French), de (German), es (Spanish)
• --provider: glm or openrouter. Auto-detected if omitted (GLM preferred when both keys present)
• --model: OpenRouter model slug (default: google/gemini-3-pro-image-preview). Ignored for GLM. See https://openrouter.ai/collections/image-models
• --size: Image dimensions, GLM only (default: 1088x1920). Valid range: 512-2048px, multiples of 32
• --output: Output directory (default: output/)
• --quality: Image quality, GLM only — "hd" or "standard" (default: hd)
• --watermark: Enable watermark, GLM only

Language Selection Rules

• Always ask explicitly. Never guess from the user's message language.
• Never default. If the user does not specify, ask again.
• Record as-typed. Pass exactly what the user said (e.g., zh, en) — do not normalize.
• Reason: GLM is a Chinese-native model; prompt language significantly affects output quality and style.

Available Sizes

• 1088x1920 (default, portrait HD)
• 1920x1088 (landscape HD)
• 1280x1280 (square)
• 1568x1056, 1056x1568
• 1472x1088, 1088x1472
• 1728x960, 960x1728

Output Format

After successful generation, display:

1. Local file path: output/<timestamp>_<prompt>.png
2. Markdown image link: ![<prompt>](<url>)

Agent Owner

This skill is executed by the main OpenClaw agent session. The generate.py script runs as a shell command via the exec tool. No sub-agents are spawned.

Success Criteria

Image generation succeeds when:

1. Script exits with code 0
2. Image file saved to output/ directory
3. Markdown image link displayed to user

Failure conditions: invalid API key, unsupported size, network timeout (120s), API quota exceeded.

Edge Cases

• Invalid size: must be 512-2048px in multiples of 32 — script will fail with API error
• Long prompts: prompt truncated to 30 chars in filename (full prompt used for generation)
• Network timeout: 120s API timeout, 60s download timeout — retry once if timeout
• Missing API key: script exits with clear error message listing search locations
• Chinese characters in prompt: supported, filename sanitized automatically

Requirements

• GLM API key configured (see Setup section above)
• Python 3 with requests package (pip install requests)